AI Image Generation and Editing
zhayujie/CowAgent
Generates or edits images from text prompts through a Python script that picks an image backend based on which API keys are configured.
Image-generation companion skill for the pumasi plugin family.
$ npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install fivetaku/gptaku-plugins-codex pumasi-image --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/fivetaku/gptaku-plugins-codex.git skills-src && mkdir -p .claude/skills && cp -r skills-src/plugins/pumasi-codex/skills/pumasi-image .claude/skills/pumasi-image && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "pumasi-image" agent skill from https://github.com/fivetaku/gptaku-plugins-codex/tree/main/plugins/pumasi-codex/skills/pumasi-image into .claude/skills/pumasi-image/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "pumasi-image", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/fivetaku/gptaku-plugins-codex/tree/main/plugins/pumasi-codex/skills/pumasi-imageType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install fivetaku/gptaku-plugins-codex pumasi-image --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/fivetaku/gptaku-plugins-codex.git skills-src && mkdir -p .agents/skills && cp -r skills-src/plugins/pumasi-codex/skills/pumasi-image .agents/skills/pumasi-image && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "pumasi-image" agent skill from https://github.com/fivetaku/gptaku-plugins-codex/tree/main/plugins/pumasi-codex/skills/pumasi-image into .agents/skills/pumasi-image/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "pumasi-image", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install fivetaku/gptaku-plugins-codex pumasi-image --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/fivetaku/gptaku-plugins-codex.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/plugins/pumasi-codex/skills/pumasi-image .cursor/skills/pumasi-image && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "pumasi-image" agent skill from https://github.com/fivetaku/gptaku-plugins-codex/tree/main/plugins/pumasi-codex/skills/pumasi-image into .cursor/skills/pumasi-image/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "pumasi-image", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/fivetaku/gptaku-plugins-codex.git --path plugins/pumasi-codex/skills/pumasi-image--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install fivetaku/gptaku-plugins-codex pumasi-image --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/fivetaku/gptaku-plugins-codex.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/plugins/pumasi-codex/skills/pumasi-image .gemini/skills/pumasi-image && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "pumasi-image" agent skill from https://github.com/fivetaku/gptaku-plugins-codex/tree/main/plugins/pumasi-codex/skills/pumasi-image into .gemini/skills/pumasi-image/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "pumasi-image", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install fivetaku/gptaku-plugins-codex pumasi-imageInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/fivetaku/gptaku-plugins-codex.git skills-src && mkdir -p .github/skills && cp -r skills-src/plugins/pumasi-codex/skills/pumasi-image .github/skills/pumasi-image && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "pumasi-image" agent skill from https://github.com/fivetaku/gptaku-plugins-codex/tree/main/plugins/pumasi-codex/skills/pumasi-image into .github/skills/pumasi-image/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "pumasi-image", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install fivetaku/gptaku-plugins-codex pumasi-image --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/fivetaku/gptaku-plugins-codex.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/plugins/pumasi-codex/skills/pumasi-image .opencode/skills/pumasi-image && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "pumasi-image" agent skill from https://github.com/fivetaku/gptaku-plugins-codex/tree/main/plugins/pumasi-codex/skills/pumasi-image into .opencode/skills/pumasi-image/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "pumasi-image", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
pumasi-imageImage-generation companion skill for the pumasi plugin family.
Pumasi Image is an agent skill from fivetaku/gptaku-plugins-codex. Image-generation companion skill for the pumasi plugin family. Use when the user asks for an image, thumbnail, poster, logo, illustration, banner, profile image, cover, icon, or any visual artifact — including casual phrasing like "그림 하나 뽑아줘", "비주얼 만들어줘", and Codex-named requests like "코덱스로 이미지 만들어줘"; this skill owns every image request. In Codex it maps to the native image generation/editing tool (gpt-image-2), with an optional Grok imagegen backend through the bundled scripts. Korean triggers — "이미지 만들어줘", "그림…
Its SKILL.md is about 3.6k tokens, which your agent loads only when the skill is triggered. The skill folder holds 11 other files, including scripts and reference files (for example `references/clarification-matrix.md`, `references/image-studio-prompt.md` and `references/keyword-mapping.md`).
It sits in Media & Creative, covering Image generation. The repository describes itself as: Codex-native GPTaku plugin marketplace. The licence is MIT.
10 steps, taken from the step headings in SKILL.md.
Read from SKILL.md and the folder at commit d3b47fc. It shows what the files ask for, not the result of running them.
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Ships 6 files in scripts/ (Shell and Python), which the agent can run.
Shell commands in SKILL.md call:
codexbashgitFrom the folder's file list and the shell code blocks in SKILL.md.
No URLs in SKILL.md. Its commands use git, which can reach the network depending on how they are called.
From URLs in SKILL.md, links to its own repository left out.
Names these keys or tokens, usually read from environment variables:
XAI_API_KEYFrom names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Pumasi Image loads about 3.6k tokens when it runs, and up to ~13k if it reads all its reference files. Until then it costs about 206 tokens; SKILL.md has 1,947 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
The automated check found no risky patterns in SKILL.md.
Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.
The full file from fivetaku/gptaku-plugins-codex at commit d3b47fc, republished under its MIT licence (© fivetaku). 1,947 words, ~3,553 tokens.
.claude/skills/pumasi-image/SKILL.md (or your agent's skills folder). This skill also uses 9 other files; get the full folder from GitHub.Codex의 네이티브 이미지 생성/편집 도구(
image_gen, gpt-image-2)로 이미지를 만든다. 코드 병렬 외주(pumasi)와 완전히 분리된 독립 스킬. 이미지 요청은 사용자가 "코덱스로"라고 말해도 전부 이 스킬이 담당한다 — pumasi로 넘기지 않는다.
references/clarification-matrix.md — 모드별 의도 파악 질문 매트릭스references/keyword-mapping.md — 비율·퀄리티 키워드 자동 매핑 + 자연어 힌트 변환표references/image-studio-prompt.md — 모드 분류 + Output Template 시스템 프롬프트 (영문 프롬프트 작성 직전에만 Read)image_gen(선택) — nanobanana 등 다른 백엔드 사용 안 함. 백엔드는 Step 3에서 사용자가 고른다(입력에 지명이 있으면 스킵). Grok을 쓸 때는 반드시 scripts/imagen.sh(--backend grok)를 경유한다(비율 강제·확장자 정합·실패 사유 표면화가 그 안에 있다). grok을 직접 부르지 말 것.api.openai.com 등 이미지 API 엔드포인트를 직접 부르는 행동은 금지. 이미지는 오직 (a) Codex 네이티브 이미지 도구 또는 (b) scripts/의 codex/grok 백엔드로만 만든다. 구독 세션이 인증을 담당하며, 이 스킬은 어떤 키도 만들거나 복사하지 않는다.{BASE_DIR}/images/{YYYY-MM-DD}/{slug}-{seq}.png목적: 구세대 diffusion 직감으로 회피 행동(HTML/CSS 분리, 텍스트 빼고 합성, 재시도 회피)을 하지 못하게 capability를 명시적으로 박는다. 모델 업데이트 시
as of날짜 기준으로 갱신.
path:로 보고한다네이티브 도구를 직접 쓰면 이 단계는 없다. scripts/imagen.sh 위임 경로를 쓸 때는 스크립트가 codex features list로 image_generation 플래그를 확인하고 꺼져 있으면 codex features enable image_generation으로 켠다 — 사용자에게는 조용히 수행한다고만 짧게 알린다.
사용자 요청에서 7가지 모드 중 하나를 결정한다:
| 모드 | 감지 키워드 |
|---|---|
| MODE_A_PORTRAIT | "프로필", "인물", "얼굴", "초상" |
| MODE_B_LANDSCAPE | "풍경", "배경", "자연", "도시", "바다", "산" |
| MODE_C_OBJECT | "제품", "물건", "아이템", "상품" |
| MODE_D_ILLUSTRATION | "일러스트", "그림", "아트", "드로잉" |
| MODE_E_THUMBNAIL | "썸네일", "커버", "대표이미지", "유튜브" |
| MODE_F_LOGO | "로고", "브랜드", "심볼", "아이콘" |
| MODE_G_CONCEPTUAL | "컨셉트", "추상", "아이디어", "상징" |
모드 판단 불확실 시 Step 3의 질문에 "모드 선택" 1개를 추가한다.
references/keyword-mapping.md를 Read하여 비율·퀄리티 자연어 힌트를 추출한다.
Codex CLI에는 객관식 카드형 질문 UI가 없다. 결정이 꼭 필요하면
$PLUGIN_ROOT/shared/questioning-policy.md §A의 채팅 번호형 선택지 블록으로 묻는다.
references/clarification-matrix.md를 Read하여 모드별 의도 파악 카테고리 3개를 확정한다.
질문 순서 (한 턴 최대 4문항 — 초과분은 우선순위 낮은 의도 질문부터 잘라낸다): 0. 백엔드 (입력에 "그록으로"/"grok"/"코덱스로"/"codex" 지명이 있으면 스킵)
질문: 어떤 백엔드로 생성할까요?
1. Codex gpt-image-2 (권장) — 임의 비율, 한글/영문 텍스트 렌더 강함
2. Grok image_gen — SuperGrok 구독 시 한계비용 0. 단 비율 9:16/16:9/1:1만, 텍스트 렌더 미검증
3. 문장으로 직접 알려주기질문 원칙:
§A 블록 예시 (단일 슬롯):
질문: 어떤 분위기로 갈까요?
1. 자동 추천 — 내용에 맞게 판단. 결정 피로 없이 진행
2. 다크 시네마틱 — 영화적, 깊은 그림자
3. 따뜻함 — 친근, 햇살, 파스텔
4. 조용한 위로감 (★) — 잔잔함, 여운
5. 문장으로 직접 알려주기references/image-studio-prompt.md를 Read하여 시스템 프롬프트를 내면화한다.
내면화 후:
프롬프트를 다음 경로에 저장 (없으면 mkdir -p):
{BASE_DIR}/.imagen/prompt-{timestamp}.md환경변수 PUMASI_IMAGE_DELEGATE_PROMPT=1 설정 시:
scripts/imagen-full.sh 호출 — 의도 한 줄 + mode + aspect + quality + 타깃 경로(+ 선택 ref 이미지)만 전달{타깃디렉토리}/.imagen-full/에 저장 (refine 복구용)기준 디렉토리 (하드코딩 금지, 동적 계산):
BASE_DIR=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
DATE=$(date +%Y-%m-%d)
TARGET_DIR="${BASE_DIR}/images/${DATE}"
mkdir -p "$TARGET_DIR"
SLUG="busan-gwangan-bridge-night" # 요청에서 계산 (핵심 명사 1~2개 → 영문 kebab-case)
SEQ=1
TARGET_PATH="${TARGET_DIR}/${SLUG}-$(printf '%02d' $SEQ).png"
while [[ -e "$TARGET_PATH" ]]; do
SEQ=$((SEQ + 1))
TARGET_PATH="${TARGET_DIR}/${SLUG}-$(printf '%02d' $SEQ).png"
done
echo "$TARGET_PATH"왜 git root 기준인가: 세션의 cwd가 항상 프로젝트 루트는 아니다. 단순 상대경로는 엉뚱한 곳에 저장될 위험이 있어, 프로젝트 루트의 images/ 하위를 기본값으로 둔다. git 저장소 밖이면 pwd 기준.
slug 예: "부산 광안대교 야경" → busan-gwangan-bridge-night, "AI 마켓플레이스 로고" → ai-marketplace-logo.
확장자는 .png. Grok 백엔드는 산출물이 보통 JPEG라 스크립트가 확장자를 맞춰 조정하므로 결과 보고는 스크립트 출력의 path:를 기준으로 한다.
기본 경로 — Codex 네이티브 이미지 도구를 직접 호출한다.
스타일 앵커(레퍼런스): 여러 장을 같은 스타일로 뽑을 때는 1장을 먼저 생성·승인받고, 나머지를 전부 그 파일을 레퍼런스로 물려 편집 도구로 생성한다(앵커 우선 패턴). 레퍼런스가 있으면 프롬프트에서 스타일 서술을 걷어내고 "첨부 이미지의 스타일·조명·색감 유지" + 피사체·구도 델타만 쓴다 — 레퍼런스가 이미 묶은 것을 긴 산문으로 재서술하면 둘이 싸운다. 스타일 일관성 수단 서열(1차 출처 리서치 2026-08-22): 레퍼런스 이미지 ≫ 텍스트 재사용 ≫ seed(어느 벤더도 프롬프트가 달라진 뒤의 재현을 보장하지 않음).
별도 codex exec 워커에 생성을 위임하거나(호스트 컨텍스트 절약), Grok 백엔드를 쓰거나, 영문 프롬프트 작성까지 위임하려면 scripts/의 결정적 래퍼를 쓴다:
# 영문 프롬프트는 호스트가 작성, 생성만 위임 (3번째 인자 비율은 실측 비교 경고용, --ref 반복 가능)
bash $PLUGIN_ROOT/skills/pumasi-image/scripts/imagen.sh \
"{prompt_file_path}" "{target_image_path}" "{aspect e.g. 16:9 — 생략 가능}" \
--backend "{codex|grok — Step 3 선택 결과, 생략 시 codex}" \
--ref "{anchor_image_path — 생략 가능}"
# 영문 프롬프트 작성까지 codex 워커에 위임 (PUMASI_IMAGE_DELEGATE_PROMPT 경로)
bash $PLUGIN_ROOT/skills/pumasi-image/scripts/imagen-full.sh \
"{intent}" "{mode}" "{aspect}" "{quality}" "{target_image_path}" "{ref_image — 생략 가능}"
# 여러 장 일괄 (partial success + per-item retry manifest; 항목에 "ref" 키로 앵커 지정 가능)
bash $PLUGIN_ROOT/skills/pumasi-image/scripts/imagen-batch.sh "{batch_json_path}"--backend grok: grok 헤드리스 image_gen으로 생성한다(--ref를 주면 image_edit). 사전 조건: grok CLI 설치($HOME/.grok/bin/grok) + grok login. 미설치/미로그인이면 스크립트가 exit 3으로 실패하니 사용자에게 알리고 codex 재생성 여부를 §A 블록으로 확인한다.
래퍼 내부 동작(codex 백엔드):
codex features list로 image_generation feature flag 재확인 (안전망)codex exec --json … < /dev/null 호출 — codex가 이미지 도구로 1장 생성. < /dev/null은 exec가 stdin EOF를 무한 대기(헤드리스 행)하는 것을 막는다. 레퍼런스는 --image=<path>(파일당 1개)로 전달 — -i FILE...은 가변 인자라 뒤따르는 프롬프트를 이미지 경로로 삼켜 "No prompt provided" 실패를 만든다(실측 2026-08-22).codex exec는 이미지를 ~/.codex/generated_images/<thread_id>/exec-*.png로 저장하고 stdout JSONL에는 base64를 싣지 않는다. 래퍼는 stdout의 thread.started.thread_id(디렉토리명과 1:1)로 이 세션 산출물만 집어 타깃에 복사한다(동시 실행과 경합 없음, 스테일 오집음 없음). 구버전 codex 호환용으로 stdout(JSONL)·세션 rollout의 base64 디코딩 폴백(extract_image.py)을 유지한다. 생성 0장이면 거짓 성공 없이 exit 5.sips) + 요청 비율과 15%↑ 괴리 시 경고. 후처리는 절대 하지 않음.HTTP_PROXY/HTTPS_PROXY)가 환경에 상속돼 있으면 codex 호출에서만 벗긴다. 프록시를 경유하면 이미지 엔드포인트 요청이 ~153초 뒤 network error로 죽는다(2026-07-23 실측: 경유 89/89 실패, 우회 시 동일 프롬프트 44초 성공). 해제는 PUMASI_IMAGE_KEEP_PROXY=1.image generation failed: … 원문을 REASON:으로 출력하고, network error면 프록시 힌트를 덧붙인다. 네트워크 실패와 정책 거부를 구분해 보고한다.⚠️ 샌드박스/승인 우회 경계 (opt-in). 래퍼는
codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox로 비대화형 실행한다 — 동작은 대상 이미지 경로 1개 쓰기로 한정된다. 신뢰하는 본인 프로젝트에서만 사용한다.
생성 모드를 판정한 후 그에 맞게 동작한다. (PNG 1장 표시 = 약 1,400~3,000 비전 토큰이 cached prefix에 박히므로 기본값은 표시 안 함.)
MODE_E_THUMBNAIL / MODE_F_LOGO / 의도에 직접 인용된 카피(따옴표)가 있을 때 — 마지막 1장만 표시file {target_image_path}로 해상도/포맷/sha1 확인 (스크립트 경로면 출력의 path: 기준)✅ 생성 완료: {path} ({해상도} {포맷}, sha1: {prefix}) + "깨졌으면 '이미지 보여줘'라고 말씀해주세요"). 이미지 표시 안 함.생성 직후 다음을 대화 컨텍스트(skill state)로 유지:
last_prompt_path: 마지막 영문 프롬프트 파일 경로 (Step 4 산출물)last_image_path: 마지막 이미지 경로 (앵커 후보)last_manifest_path: 마지막 manifest.json 경로 (imagen-full.sh 사용 시)리파인 판정:
last_prompt_path Read + 사용자 델타만 patchlast_image_path 표시 (자동 review 모드)image-studio-prompt.md는 절대 재로드 X (28KB)last_prompt_path + 델타 patch만. image-studio-prompt.md는 절대 재로드 X.PUMASI_IMAGE_DELEGATE_PROMPT=1은 feature flag로만 활성화. Step 7 fast 모드와 동시 적용 시 MODE_REFINE 컨텍스트 부서짐 위험.| 구분 | pumasi (코드) | pumasi-image (이미지) |
|---|---|---|
| 스킬 디렉토리 | skills/pumasi/ | skills/pumasi-image/ |
| 자동 트리거 | "구현", "개발", "기능", "코드" | "이미지", "그림", "썸네일", "로고" (Codex 지명 포함) |
| 백엔드 | CLI 워커 (Codex 멀티에이전트/codex exec/grok/cursor…) | Codex 네이티브 이미지 도구 (+ 선택 Grok image_gen) |
| 작업 dir | .pumasi/ | 없음 (단발 요청) |
두 스킬은 같은 플러그인 안의 독립 모듈이며 서로 간섭하지 않는다.
references/image-studio-prompt.md — 모드 분류 + Output Template 시스템 프롬프트 (Format/Medium Guard 포함)references/clarification-matrix.md — 모드별 의도 파악 질문 매트릭스references/keyword-mapping.md — 비율·퀄리티 키워드 자동 매핑 + 자연어 힌트 변환표scripts/imagen.sh — feature flag 확인·활성화 + codex exec --json 호출(또는 --backend grok) + thread_id 기반 산출물 회수(+base64 폴백) + 프록시 우회 + 실패 사유 표면화 + --ref 스타일 앵커 + SHA1/해상도 검증scripts/imagen-full.sh — 영문 프롬프트 작성까지 codex 워커에 위임(manifest/prompt/log 보존). 프록시 우회·실패 사유 표면화 동일 적용scripts/imagen-batch.sh — 여러 장 일괄(partial success + per-item retry manifest, 항목별 ref)scripts/imagen-cleanup.sh — ~/.codex/generated_images/ 누적 정리(기본 DRY-RUN, --apply 시 trash)scripts/extract_image.py — 구버전 codex 호환: JSONL/세션 rollout의 image_generation_call base64를 구조 검증 후 PNG로 저장scripts/test-imagen-capture.sh — 회수 계약 회귀 테스트(mock codex/grok, 32 assertions)codex features 서브커맨드 사용 가능$HOME/.grok/bin/grok) + grok login (구독 세션, XAI_API_KEY 불필요)© fivetaku, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
SKILL.md and 9 other files (scripts, references) in plugins/pumasi-codex/skills/pumasi-image of fivetaku/gptaku-plugins-codex.
Open the folder on GitHubat commit d3b47fc
Pumasi Image next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Pumasi Image this skillfivetaku/gptaku-plugins-codex | 128 | — | ~3.6k | Automated safety check: Pass | MIT | |
| AI Image Generation and Editingzhayujie/CowAgent | 47k | — | ~1.3k | Automated safety check: Pass | MIT | |
| Structured Image Generationbytedance/deer-flow | 84k | 4 repos | ~2.9k | Automated safety check: Pass | MIT | |
| Canghe Comicfreestylefly/canghe-skills | 461 | 8 repos | ~3.2k | Automated safety check: Pass | None | |
| Generate Imageynulihao/AgentSkillOS | 618 | 10 repos | ~1.7k | Automated safety check: Notes | None | |
| GPT Image Generation CLIwuyoscar/GPT-Image2-Skill | 5.7k | — | ~2.5k | Automated safety check: Notes | MIT |
zhayujie/CowAgent
Generates or edits images from text prompts through a Python script that picks an image backend based on which API keys are configured.
bytedance/deer-flow
Turns an image request into a structured JSON prompt and runs a bundled Python script to generate the picture, optionally guided by reference images.
freestylefly/canghe-skills
Knowledge comic creator supporting multiple art styles and tones.
ynulihao/AgentSkillOS
Generate or edit images using AI models (FLUX, Gemini). An agent skill from ynulihao/AgentSkillOS.
wuyoscar/GPT-Image2-Skill
Generates and edits images with GPT Image 2 or 2.5 through a packaged CLI and a prompt gallery, after settling which model fits the request.
LiamGvchi/gc-minimal-zine-poster
Creates or analyzes quiet, paper-texture zine posters with big negative space, one color accent and experimental type, returning an image prompt and the generated poster.
fivetaku/gptaku-plugins-codex
Runs a multi-agent deep research workflow in seven phases, from scoping questions to a final report with source triangulation, state tracking and quality ratings.
fivetaku/gptaku-plugins-codex
This skill should be used when the user asks to "스킬 만들어줘", "에이전트 만들어줘", "커맨드 만들어줘", "스킬러들의 수다", "수다", "Codex 스킬 만들어줘", "이 스킬 분석해줘", "이 스킬 개선해줘", "skill builder", "make a skill", "create a skill"…
fivetaku/gptaku-plugins-codex
A skill your agent uses when the user runs /dd or /ㅇㅇ (Hangul IME alias — typing "dd" in Korean IME produces "ㅇㅇ") to act on the current OS clipboard (text or image) without pasting it into chat.
fivetaku/gptaku-plugins-codex
Adaptive access for blocked websites — tries every method until one works.
fivetaku/gptaku-plugins-codex
Fetch and explain official documentation for any library, framework, API, or service using an llms.txt-first strategy — triggers on "How do I…", "What is…", "How does X work", "Best practice for…"…
fivetaku/gptaku-plugins-codex
Explain Git and GitHub concepts using cloud-folder analogies for non-developers, and carry the cross-cutting teaching principles for the whole git-teacher skill set.
Categories
Image-generation companion skill for the pumasi plugin family. Pumasi Image is an agent skill from fivetaku/gptaku-plugins-codex. Image-generation companion skill for the pumasi plugin family.
Pumasi Image fits situations like: the user asks for an image; any visual artifact — including casual phrasing like 그림 하나 뽑아줘; Codex-named requests like 코덱스로 이미지 만들어줘; this skill owns every image request.
Run `npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a claude-code`. Or copy the skill folder (plugins/pumasi-codex/skills/pumasi-image in fivetaku/gptaku-plugins-codex) into .claude/skills/pumasi-image in your project. Claude Code loads it when a task matches its description.
Run `npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a codex`. Or copy the skill folder (plugins/pumasi-codex/skills/pumasi-image in fivetaku/gptaku-plugins-codex) into .agents/skills/pumasi-image in your project. Codex loads it when a task matches its description.
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add fivetaku/gptaku-plugins-codex --skill pumasi-image -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/pumasi-image, .gemini/skills/pumasi-image, .github/skills/pumasi-image and .opencode/skills/pumasi-image in your project.
Going by SKILL.md and its folder, Pumasi Image needs a shell and Python for the scripts in its folder, the command-line tools its instructions call (codex, bash and git) and credentials named XAI_API_KEY. Our summary lists: Python 3; A Bash shell.
SKILL.md contains no URLs. Its commands use git, which can reach the network depending on how they are called. This is read from the text; nothing was executed.
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.
Pumasi Image is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.
About 3.6k tokens (SKILL.md is roughly 14k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 9.8k tokens, read only when the agent opens those files.
Skills that share tags, products or a category with Pumasi Image: AI Image Generation and Editing (zhayujie/CowAgent, 47k stars), Structured Image Generation (bytedance/deer-flow, 84k stars), Canghe Comic (freestylefly/canghe-skills, 461 stars) and Generate Image (ynulihao/AgentSkillOS, 618 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
fivetaku (a GitHub user) maintains it in fivetaku/gptaku-plugins-codex, which has 128 GitHub stars. The repository holds 25 skills in this directory. The repository was last updated on September 8, 2026.
Source: fivetaku/gptaku-plugins-codex on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.